Local voice-to-text for macOS
100% offline • Apple Silicon optimized • Menu bar app
A macOS menu bar app for local speech-to-text powered by WhisperKit. Press a hotkey, speak, and text appears in any app — no internet required.
- Download the latest
.dmgfrom GitHub Releases - Open the DMG and drag LocalWhisper to your Applications folder
- Open LocalWhisper from Applications
- Grant Microphone and Accessibility permissions when prompted
Note: On first launch, you may see "unidentified developer" warning. Right-click the app and select "Open" to bypass this.
git clone https://github.com/t2o2/local-whisper.git
cd local-whisper
swift build && swift run- Grant Microphone and Accessibility permissions when prompted
- Hold your shortcut key (default:
Ctrl+Shift+Space) to start recording - Speak while holding the key
- Release to stop recording and transcribe
Text is automatically typed into your focused app.
- 🎤 Global Hotkey — Hold to record, release to transcribe (default:
Ctrl+Shift+Space) - 🔒 100% Offline — All processing on-device, no data leaves your Mac
- ⚡ Fast — CoreML + Neural Engine acceleration on Apple Silicon
- 📝 Auto-inject — Transcribed text typed directly into focused field
- 📖 Custom Dictionary — Add words/names for accurate transcription of technical terms, proper nouns, etc.
- macOS 14.0+ (Sonoma)
- Apple Silicon (M1/M2/M3/M4)
- 8GB RAM minimum (16GB+ for large models)
Click the menu bar icon to:
- Change keyboard shortcut
- Select transcription model (tiny → large-v3)
- Add custom vocabulary (product names, technical terms, proper nouns)
- Adjust settings
Add words you want transcribed correctly in Settings → Custom Vocabulary. This helps the model recognize:
- Product names (e.g., "WhisperKit", "CoreML")
- Technical terms (e.g., "Kubernetes", "PostgreSQL")
- Proper nouns (e.g., names of people, places, companies)
Tip: Works best with larger models (small, medium, large-v3). The dictionary provides spelling hints, not instructions.
- Model Guide — Model comparison, benchmarks, recommendations
- Architecture — Project structure, development guide
All transcription happens locally. No audio is sent over the network. No analytics or telemetry.
MIT
- WhisperKit — Swift Whisper with CoreML
- KeyboardShortcuts — Global hotkeys
- OpenAI Whisper — Original model